[tied] Re: pre-Nostratic *male[:]k?xa, 'milk (vb.)'

From: Richard Wordingham
Message: 43407
Date: 2006-02-13

--- In cybalist@yahoogroups.com, "Brian M. Scott" <BMScott@...> wrote:
>
> At 9:30:27 PM on Sunday, February 12, 2006, mkelkar2003
> wrote:
>
> > --- In cybalist@yahoogroups.com, "Brian M. Scott" <BMScott@> wrote:
>
> >> At 6:21:19 PM on Sunday, February 12, 2006, mkelkar2003
> >> wrote:
>
> [...]
>
> >>> Of the 35 most basic word vocabulary PIE and OC have 23%
> >>> cognates. That is too many for a chance occurence.
>
> >> This is a non sequitur. It's also obviously incorrect:
> >> apart from a tiny handful of possible borrowings, there are
> >> *no* known cognates, since OC is not known to be related to
> >> PIE at all.

Let us remember that the chance similarity when applying comparisons
by Swadesh's rules is about 8% For 35 meanings, that means an average
of 2.8 matches could be due to chance! 23% of 35 means 8 items matched.

The probability of getting 8 or more matches out of the 35 is about
0.8%, which is impressive. However, that is the *best* of 5
comparisons. It's not too many for it be mere chance, but I'm not
sure how many comparisons were done to get a good score. The problem,
as my stats lecturer said, is that 'all samples are peculiar'. Theory
says you form the hypothesis before you look at the data; practice
says you get the data first. Unfortunately, in historical sciences
you often cannot gather a fresh data set to test an idea formed by
looking at a set of data.

> > Refer to Table 2 below:
>
> > http://www.ee.cuhk.edu.hk/~wsywang/publications/lg_diversity.pdf
>
> >> Presumably you mean that 23 items are superficially
> >> similar.
>
> It would be nice if you'd stop wasting my time: the caption
> over Table 2 makes it clear that the table refers to items
> that are superficially similar, not to demonstrable
> cognates.

That's because there aren't enough possible cognates to go very far
and confirm the correspondences. The idea is that the 35 meanings are
those that keep their primary words best. Most of the problems that
Mark Rosenfelder refers to at www.zompist.com do not apply to this
comparison. However, the title of 'Apparent Cognates' is immediately
justified by the method - you could very well be looking at 5 cognates
and 3 coincidences even if the link between IE and Sino-Tibetan is valid.

However, this is getting off-topic, and should be taken to Nostratic-l
or even further afield, e.g.
http://groups.yahoo.com/group/Macro-Familes-L . Prompt, justified
objections to my mathematics are acceptable, though.

Richard.